User Activity Analysis Via Network Traffic Using DNN and Optimized Federated Learning based Privacy Preserving Method in Mobile Wireless Networks

Mobile and wireless networking infrastructures are facing unprecedented loads due to increasing apps and services on mobiles. Hence, 5G systems have been developed to maximise mobile user experiences as they can accommodate large volumes of traffics with extractions of fine-grained data while offering flexible network resource controls. Potential solutions for managing networks and their security using network traffic are based on UAA (User Activity Analysis). DLTs (Deep Learning Techniques) have been recently used in network traffic analysis for better performances. These previously suggested techniques for network traffic analysis typically need voluminous information on network usages. Hence, this work proposes OFedeMWOUAA (optimal federated learning-based UAA technique with Meadow Wolf Optimisation) and DNN (deep Neuron Networks) for minimizing risks of data leakages in MWNs (Mobile Wireless Networks). In the proposed OFedeMWOUAA, the need to submit data to cloud servers does not arise because it trains DLTs locally and only uploads model gradients or knowledge weights. The OFedeMWOUAA approach effectively decreases dangers to data privacies with very minor performance losses in simulations.